In the logic called for curl_multi_fdset().
File descriptors larger than FD_SETSIZE size are simply ignored, which
of course will make things break but at least it does not trash memory.
Reported-by: Stanislav Fort (Aisle Research)
Closes#19439
To debug the Windows CI fails further. Acting on the suspicions that
`taskkill` may sometimes be applied to the wrong process.
- log task info, and task child info before calling `taskkill` on a PID.
(on native Windows.)
One of the calls needs PowerShell.
- add env `CURL_TEST_NO_TASKKILL` to disable using `taskkill`.
- add env `CURL_TEST_NO_TASKKILL_TREE` to use `taskkill` without
`-t`, meaning to kill the process, but not child processes.
- GHA/windows: disable `taskkill` calls, to see what happens.
I'll revert or tweak this in a future commit depending on results.
Ref: https://github.com/curl/curl/discussions/14854#discussioncomment-13062859
Ref: https://github.com/curl/curl/discussions/14854#discussioncomment-14913014Closes#19421
Before this patch servers were loading the original data source file
(from `tests/data/test*`) if they failed to open the preprocessed data
file.
It was causing issues in many (most?) tests, because original data files
are not preprocessed, thus may be incomplete and/or come with wrong
newline characters. It's also causing difficult to diagnose issues when
a test accidentally references another test's data, which by chance
makes the test pass initially, until either that or the executed test
data gets an update, and breaking it, as seen in #19329.
Historically, the fallback existed first, then the preprocessed copy.
The fallback is no longer used by tests (except by stray accidents).
Fix it by dropping the fallback logic and relying on the preprocessed
data file saved there by the runtests framework.
Also fix two remaining test data cross-references:
- test1565: reference own server input data instead of test1's.
- test3014: reference own server input data instead of test1439's.
Ref: #19398
Follow-up to aaf9522a2c#19329Closes#19429
If CMake reports the target platform not supporting shared libs, turn
`BUILD_SHARED_LIBS` off by default. CMake 3.30+ fails with an error
when trying to create a `SHARED` target for such platforms. Earlier
versions used a workaround that may or may not have worked in practice.
Ref: https://cmake.org/cmake/help/v3.30/policy/CMP0164.html
Seen this with a build setting `-DCMAKE_SYSTEM_NAME=Generic`, e.g.
AmigaOS.
Note this may introduce incompatibility for "Generic" targets, which
support shared libs. If that's the case, set `BUILD_SHARED_LIBS=ON`
manually.
Also drop AmigaOS-specific logic handled automatically after this patch.
Ref: https://cmake.org/cmake/help/v3.7/command/get_property.html
Ref: https://cmake.org/cmake/help/v3.7/prop_gbl/TARGET_SUPPORTS_SHARED_LIBS.htmlCloses#19420
They were disabled since these jobs ran in Zuul. The tests are 320, 321,
322, 323, 324. Of which, 323 runs in CI, the rest needs `gnutls-serv`
with SRP enabled, which is not available in current jobs and no longer
offered by Ubuntu's `gnutls-bin` package. 324 doesn't appear as
a skipped test, 323 seems to be running fine, the rest are logged as
skipped. This suggests it's safe to drop the exceptions.
Closes#19413
Build and cache LibreSSL locally with Fil-C and enable it in the Fil-C
job.
Also:
- disable test 776 in the Fil-C job. It fails consistently, and due to
flakiness seen earlier its result is disabled. In this job it seems to
be adding 1 to 9 minues to the test run step and fails consistently.
- include Fil-C version in the LibreSSL cache key to prepare for Fil-C
ABI changes.
- GHA/linux: fully quote `tflags` values to avoid breaking YAML.
Tested and confirmed working with OpenSSL too, but ended up with
LibreSSL for faster, smaller builds.
Closes#19407
To allow running pytests on more curl configurations.
Also delete a redundant H3 feature check from test_17_14_expired_cert.
Cherry-picked from #19407Closes#19412
The short descriptions describe the data each info retrieves. The info
itself does not 'get' the data.
This simplifies and shortens the descriptions and make them more
consistent.
Closes#19406
A regression in curl 8.17.0 led to a customer CAPATH set by the
application (or the curl command) to be ignored unless licurl was built
with a default CAPATH.
Add test cases using `--capath` on the custom pytest CA, generated with
the help of the openssl command when available.
Fixes#19401
Reported-by: Brad King
Closes#19308
Requirements for Fil-C:
- not to accidentally pick up system headers. E.g. from `/usr/include`
on Linux. It can happen when any dependency is auto-detected on this
header path. This makes Fil-C find the wrong system headers, which
in turn breaks the configuration step in subtle ways (with CMake) and
less subtle ways (autotools). Then CMake ends up running into an error
while compiling.
- build all dependencies with Fil-C too.
(this patch doesn't build any dependencies yet.)
- "unity" mode disabled. It should work, but needs a lot of memory and
slower than a standard compiler, or a Fil-C non-unity build.
- x86_64 Linux host platform when using the pre-built toolchain.
Observations on a minimal, static build made with no dependencies and
Fil-C 0.674 (based on clang 20.1.8).
- curl tool sizes:
- cmake, default, w/o -O: 30 MB (gcc 14.2.0: 1.7 MB)
- cmake, default, w/o -O, stripped: 29.6 MB (gcc: 1.4 MB)
- cmake, Release, -O3: 7.2 MB (gcc: 1 MB)
- cmake, Release, -O3, stripped: 6.8 MB (gcc: 0.93 MB)
- autotools, default, -O2: 7 MB
- libcurl.a size is 32 MB (cmake, default, w/o -O) (gcc: 2.7 MB)
- build times 3-3.5x longer (compared to system gcc 14.2.0):
- all runtests available pass OK.
- all pytests skipped due to missing features/dependencies.
- shared libcurl builds also work (cmake, default: 25 MB libcurl.so and
5.75 MB (5.6 stripped) curl tool)
- autotools works fine too, with dependencies disabled or set to avoid
`/usr/include`.
Closes#19391
On some platforms, socket descriptors may use types larger than int.
When these values exceed INT_MAX, conversion to int can truncate to
negative values causing RTMP connection failures, and even accidentally
affect other socket when high-value descriptors map to existing
lower-value sockets after integer conversion. This check ensures socket
values are within the safe range before passing them to the RTMP
library.
Closes#19399
Resumed TLS sessions skip OCSP stapled-response verification. Force a
full handshake so verifystatus() runs.
Follow-up to 4bfd7a9615
Pointed out by ZeroPath
Some internal functions always return CURLE_OK.
- Curl_http_proxy_get_destination() does that from bb4032a, (2 years
ago) And the original inline code does not need to check the status.
- Curl_wildcard_init() does that from e60fe20. (8 years ago)
- Curl_initinfo() does that from a very beginning.
- Curl_pgrsSetDownloadCounter() did not have a return before 914e49b,
ad051e1 recovered its content (2 years ago) but did not completely
recovered the changes related to it.
- auth_digest_get_qop_values() does that from 676de7f.
This directly changes their type to void and cleaned the remaining
checks for their return value.
Closes#19386
Instead of happily ignoring return codes.
Calls that allocate data, like duplicating strings, can fail because of
lack of memory which could then leave the option unset and curl would
unknowingly continue (if the memory shortage was momentary).
Closes#19385
When Curl_conn_cf_recv() returns error, the variable might not be
assigned and the tracing output may (harmlessly) use it uninitialized.
Also add a comment about the typecast from size_t to int being fine.
Pointed out by ZeroPath
Closes#19393
Both CURLOPT_PROTOCOLS_STR and CURLOPT_REDIR_PROTOCOLS_STR would
previously return error on bad input but would wrongly still store and
keep the partial (unacceptable) result in the handle.
Closes#19389
curl is for transfers so disabling all protocols has to be a mistake.
Previously it would allow this to get set (even if curl_easy_setopt()
returns an error for it) and then let libcurl return error instead.
Updated 1474 accordingly.
Closes#19388
There is no more mixed-newline file in the repository after this patch.
Except for`.bat` and `.sln` files (4 in total), all files use LF
newlines.
Also:
- `spacecheck.pl`: drop mixed-EOL exception for test data.
- runtests: add option `-w` to check if test data has stray CR bytes in
them.
- build: enable the option above in test targets, except the CI-specific
one where `spacecheck.pl` does this job already.
- tested OK (with expected failures) in CI with stray CRs added.
- cmake: enable option `-a` for the `tests` target. To continue testing
after a failed test.
Follow-up to 63e9721b63#19313
Follow-up to 6cf3d7b1b1#19318
Follow-up to 4d2a05d3fe#19284Closes#19347
With `check_finished()` triggered by notifications now, the
`progress_meter()` was no longer called at regular intervals.
Move `progress_meter()` out of `check_finished()` into the perform loop
and event callbacks.
Closes#19383
Syncing behavior with `CURL_CA_BUNDLE` and autotools.
`/etc/ssl/certs` is empty by default on macOS systems, thus no likely
auto-detection finds something there.
Follow-up to eefd03c572#18703Closes#19380
With GnuTLS 3.8.0+ the build-time SRP feature detection always succeeds.
It's also disabled by default in these GnuTLS versions.
When using TLS-SRP without it being available in GnuTLS, report
the correct error code `CURLE_NOT_BUILT_IN`, replacing the out of memory
error reported before this patch.
Also add comments to autotools and cmake scripts about this feature
detection property.
Detecting it at build-time would need to run code which doesn't work
in cross-builds. Once curl requires 3.8.0 as minimum, the build-time
checks can be deleted.
```
# before:
curl: (27) gnutls_srp_allocate_client_cred() failed: An unimplemented or disabled feature has been requested.
# after:
curl: (4) GnuTLS: TLS-SRP support not built in: An unimplemented or disabled feature has been requested.
```
Ref: dab063fca2
Ref: a21e89edacCloses#19365
